Files
accounted/extensions/general/mcp-server
Jakob WennbergandClaude Fable 5 816b1769c8 feat(dimensions): PR6 retro-tagging — audited retag carve-out, BulkTagWorkbench, staged MCP tool (#867)
* feat(dimensions): PR6 retro-tagging — audited retag carve-out, workbench, staged MCP tool

Tier-2 retro-tagging (founder decision №1, approved 2026-07-02): posted
entries in OPEN periods can have their dimension tags changed through ONE
audited path — everything about the verifikat itself stays immutable.

Carve-out (migration 20260702170000): the line-immutability trigger gains a
single narrow branch — while the transaction-local GUC set by the RPC is
active, an UPDATE of a posted line is admitted iff every non-dimension
column is unchanged, enforced by a whole-row to_jsonb diff (any future
column is protected by construction; mirrors cost_center/project are in the
changeable set because they are derived views of dimensions['1']/['6']).
Precedent: mark_entry_as_opening_balance (20260613120000).

retag_line_dimensions RPC: tenant guard (20260619130100 pattern), writer
gate (viewers rejected), posted-only, open period + company lock date
enforced, every code validated against the ACTIVE registry, immutable
dimension_retag_log row (before/after/actor/reason, INSERT-only via its own
trigger, no FKs so the trail survives hard-deletes) written BEFORE the
carve-out UPDATE. Idempotent no-op without a log row. Untag ({}) supported.
Legal position per the plan: dimensions are internredovisning metadata, not
BFL 5 kap 7§ verifikat content — this is strictly more conservative than
Fortnox/Visma (dimension-only diffs, open periods only, immutable log,
storno past locks — Tier 3 has no exceptions).

Mandatory pg suite (11 tests): GUC-less updates still blocked; amounts/
description can never change even under the GUC (transaction-local);
closed/locked/lock-date, role, registry, draft and cross-tenant rejections;
log immutability; gnubok.allow_delete bulk path unaffected.

UX (all writes through the ONE RPC): pencil on posted-voucher lines in
bookkeeping/[id] ("Påverkar endast internredovisningen, inte verifikatet")
+ retag-history card; BulkTagWorkbench at /dimensions/tagging (filters,
shift-select, merge vs "Ersätt tagg" replace mode, reversal-pair warning
with "Inkludera motverifikat" auto-selection, per-line failure display).

MCP: gnubok_tag_journal_lines (bookkeeping:write) — filter block resolved
via resolve-don't-select, ≤500 lines, staged via pending_operations (new
op type migration 20260702171000, medium risk tier, shared Zod validation
boundary between staging and commit; executor loops the RPC per line with
partial-success aggregation).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(dimensions): address #867 review — SQLSTATE classification, blocking storno confirm, documented divergence

- Retag route classifies RPC errors by SQLSTATE instead of message-regex:
  P0001 (every rule violation in the RPC) → 409 verbatim, 42501 (tenant
  guard) → 403, anything else → logged 500 with a generic message. No more
  substring sniffing.
- The workbench's storno-pair warning escalates to a BLOCKING confirmation
  naming the unselected counter-vouchers before apply (Srf U 14 gross
  reporting — one-legged retags silently skew project P&L; the banner alone
  was advisory).
- The empty-bag divergence is now documented on both schemas as intentional:
  the direct dialog/workbench path allows {} (human untags phantom codes,
  logged with reason), the MCP staged path rejects it (agents never
  bulk-clear history).

Triage notes: the log's missing FKs are the point (behandlingshistorik must
survive undo_sie_import hard-deletes — a cascade would erase the trail);
SIE exports are generated fresh on demand, never cached, so post-retag
exports carry the new object lists automatically; date-scoped registry
values are deliberately not enforced at retag because entry creation does
not enforce them either — enforcing in one path only would be incoherent
(both belong to the PR10 rules engine).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-02 17:02:34 +02:00
..
2026-05-04 11:12:29 +02:00

Accounted MCP server

JSON-RPC 2.0 server exposing the Accounted bookkeeping engine to MCP clients (Claude Desktop, Claude Code, etc.). Endpoint: /api/extensions/ext/mcp-server/mcp. OAuth and stdio bridge live alongside the API surface — see app/api/mcp-oauth/ and packages/gnubok-mcp/.

Tool authoring contract

Enforced by tests in __tests__/ — these are not style preferences, they're guard rails.

  1. additionalProperties: false on every inputSchema. Guarded by strict-schemas.test.ts. Forces clear rejections on hallucinated fields instead of silent ignores.
  2. Descriptions ≤ 280 chars. Guarded by output-schema.test.ts. No Args: / Returns: / Examples: prose — those belong in JSON Schema. Use agent-native hints ("Use to…", "Call X first", "HIGH risk").
  3. Staged-operation envelope for write tools — outputSchema: STAGED_OPERATION_SCHEMA (server.ts). Fields: staged, risk_level, actor, message, preview, period_status?, next?. The staged: true boolean is the explicit completion signal; agents must not infer completion from prose. Do NOT introduce a parallel { success, shouldContinue, output } envelope.
  4. period_status threading — any tool that ties to a fiscal-period-bound date (categorize, mark paid, create voucher, correct/reverse entry, approve supplier invoice) passes dateForPeriodCheck to stagePendingOperation. Response then includes period_status: { period_id, status: open|locked|closed, lock_date } so widgets and agents disable writes without round-trips.
  5. Scope mapping — every new tool needs an entry in lib/auth/api-keys.ts TOOL_SCOPE_MAP. Missing entries default to deny.
  6. Tests for new write tools — add staging-gate coverage to __tests__/voucher-tools.test.ts (or a sibling) plus executor coverage to lib/pending-operations/__tests__/voucher-executors.test.ts if the tool stages a new operation_type.

Determinism / cache stability

Tool definitions (name, description, inputSchema, outputSchema, annotations) are declared as static object literals at module load — no timestamps, no UUIDs, no Date/Math.random in the definition layer. This makes the tools/list JSON payload byte-stable across requests, which lets agent-side prompt caches stay warm. Do not introduce per-request non-determinism into the definitions block. Anything time-bound or random belongs inside execute().

For internal Anthropic API usage (today only extensions/general/invoice-inbox/lib/extract-invoice-fields.ts): annotate stable prefixes with cache_control: { type: 'ephemeral' } and log usage.cache_read_input_tokens for hit-ratio observability. The 1h TTL from the agent-native API plan (item 10) requires the direct Anthropic API; Accounted's Bedrock path defaults to a shorter TTL.

Payload-size watchdog

payload-size.bench.test.ts enforces a tools/list JSON payload ceiling (currently 36,000 tokens — bumped from 32,000 when top-level Tool.title landed on all tools for Claude Connectors Directory readiness). If the test fires, the right answer is rarely "raise the ceiling" — instead, trim descriptions or leverage gnubok_search_tools (already deployed; tool definitions can defer to it for discovery rather than enumerating in tools/list).

Where things live

  • server.ts — the tools array + JSON-RPC dispatcher
  • tool-result.ts — withNext(), toToolError() response helpers
  • resources/ — read-only Accounted:// URIs (active company, period, recent activity, capabilities, attention items, voucher gaps, chart of accounts, VAT treatments)
  • widgets/ — inline HTML widgets (receipt-matcher, vat-review)
  • prompts/ — slash-command-style prompts
  • skills/ — domain-knowledge skill bodies served via gnubok_load_skill
  • __tests__/ — strictness guards + per-tool coverage